Artificial Intelligence Interview Questions and Answers

How would you define artificial intelligence?

Building intelligent machines that can mimic human behaviour is the goal of the computer science subfield known as artificial intelligence. In this context, an intelligent machine is one that has decision-making abilities and can behave and process information like a human. The compound word "artificial intelligence" refers to "man-made thinking ability."

Instead of having to pre-program a computer to perform a task, we can create one with which was before algorithms that could operate independently using artificial intelligence.

What justifies the use of artificial intelligence?

Artificial intelligence aims to build intelligent machines that can mimic human behaviour. Artificial intelligence (AI) was needed in today's world to solve complex issues, enhance daily life by automating repetitive tasks, save labour, and perform a variety of other tasks.

What varieties of AI are there?

Based on its features and capabilities, artificial intelligence can indeed be classified into many forms.

Depending on abilities:

Narrow AI, often known as weak AI, is only capable of carrying out a few specific tasks intelligently. A weak AI example is Siri.

Generic AI: Intelligent machines that are as effective as humans at doing any intellectual endeavour.

Strong AI is the speculative idea of a machine which will be smarter than humans and outperform them in intelligence.

based on capabilities:

Reactive Machines: The first forms of AI are solely reactive machines. They are present-oriented and unable to retain past activities. Consider Deep Blue.

Limited Memory: Because its names imply, it has a finite amount of space for storing historical information or experience. An illustration of this kind of AI is the self-driving automobile.

Theory of Mind: This sophisticated AI is able to comprehend human emotions, people, and other things in the actual world.

List of actual-world uses for AI.

AI has several practical applications, some of which are listed below:

Google Search Engine: Thanks to various AI technologies, as soon as we start typing something into the search engine, Google quickly provides us with recommendations that are pertinent to our query.

Apps for ridesharing: Uber and other ride-sharing services utilise machine learning and artificial intelligence to figure out the kind of transportation, how long it will take to arrive once the user hails it, how much it will cost, and other factors.

Email spam filters are also implemented using artificial intelligence (AI), allowing you to only receive critical and pertinent emails in your inbox. Gmail successfully removes 99.9% of spam emails, according to research.

What kinds of machine learning are there?

Three basic categories can be used to classify machine learning:

Learning algorithm is a sort of machine learning whereby a computer requires outside supervision in order to learn from data. The supervised machine learning models are trained using the labelled dataset. The two basic issues that can be resolved with machine learning under supervision are regression and classification.

Unsupervised Learning: This term refers to a sort of machine learning that occurs when the machine can learn from the data on its own without any external supervision. The unlabelled dataset can be used to train the unsupervised models. They are employed in order to address the Association & Clustering issues.

Reinforcement learning is a learning process where an agent communicates with its surroundings by taking actions and learns through feedback. The agent receives input in the way of rewards; for example, he receives a positive reward for each good activity and a negative reward for each bad action. The agent is not under any oversight. Reinforcement learning makes use of the Q-Learning algorithm.

What does "Q-Learning" mean?

A well-liked algorithm for reinforcement learning is Q-learning. The Bellman equation is its foundation. In this method, the agent seeks to discover the rules that can offer the optimal courses of action for maximising rewards under specific conditions. These ideal strategies are acquired by the agent through prior knowledge.

In Q-learning, the agent aims to maximise the value of Q, which is used to demonstrate the caliber of the activities at each stage.

How is deep learning applied in the real world?

A subtype of machine learning called "deep learning" imitates how the human brain functions. It is modelled after the neurons found in the human brain and uses neural networks to tackle challenging real-world issues. Deep neural learning or the deep neural network system are other names for it.

Where are intelligent agents utilised and what do they do in AI?

Any independent entity that uses sensors to detect its surroundings and actuators to act on it might be considered an intelligent agent. These intelligent agents are used in the following AI applications:

Accessing information repeatedly and using navigational tools like search engines Chatbots, domain experts, etc.

How is AI connected to machine learning?

Machine learning is a subfield and subset of artificial intelligence. It is a technique for getting AI. As these are distinct ideas, the relationship between them might be stated as "AI uses many machine learning concepts and techniques to solve complicated issues."

What Is the Markov Decision Process?

A reinforcement learning issue can be resolved using the Markov decision-making method, also known as MDP. As a result, MDP is used to formalise the RL problem. One way to think of it is as a mathematical response to a problem related to reinforcement learning. By selecting the best policy, the major goal of this step is to reap the most beneficial benefits possible.

The agent moves from starts Turning to state S2 as well as from the starting position to the final outcome in this process by doing action A, and while performing these acts, the agent receives rewards. The agent's course of action is referred to as the policy.

What does "reward maximisation" mean to you?

The phrase "reward maximisation" is used to characterise one of the goals of the training data agent in reinforcement learning. In the actual world, a recompense is a positive compliment you get when you do something to transform one state into the other. The agent is rewarded for their good performance. when using optimal policies, and a reward is taken away for poor actions. By using the best policies, the agent aims to maximise these benefits, a process known as reward maximisation.

What do the terms "parametric" and "non-parametric" mean exactly?

Machine learning models can be classified as either parametric or non-parametric. In this situation, the parameters utilised to build the machine learning model are referred to as parameters. A description of these models is provided here:

Deep - learning (ML) models are generated using parametric models, which use a set number of parameters. Strong data hypotheses are taken into consideration. Examples for parametric models include multiple regression, logistic regression, naïve bayes, perceptrons, etc.

The non-parametric model uses a range of parameter values. It considers a few data-related assumptions. These models are unknown and perform well with larger datasets. The Decision Tree, K-Nearest Neighbor, SVM with Gaussian kernels, and other non-parametric models are examples.

How can the Hidden Markov model work?

A statistical model known as the hidden Markov model is used to depict the probability distributions across a series of data. Markov defines that all assumes that now the process meets the Markov property, and hidden defines that it implies that the condition of a system generated at a certain time is concealed from the observer. Most temporal data is used using HMM models.

How does Strong AI vary from Weak AI, and what is Strong AI?

Strong AI: Real intelligence, or artificial intelligence produced by people with feelings, self-awareness, and emotions akin to those of humans, is what strong AI is all about. The idea of creating AI beings with human-like thinking, reasoning, & decision-making ability is currently just an assumption.

Weak AI: The present level of artificial intelligence development, weak AI is concerned with building intelligent robots and agents that can assist people and resolve challenging challenges in the real world. Weak AI algorithms include Siri and Alexa, as examples.

Provide a succinct overview of the AI Turing test?

The Turing test is one of the most popular cognitive tests for artificial intelligence. Alan Turing created the Turing test in the year 1950. A machine's ability to think like something a human is being tested in this experiment. This test states that a computer could only be considered intelligent when it's able to imitate human behaviour in specific situations.

Three people are involved in this exam: a computer, a test subject, and a test interrogator. The interrogator's job is to determine which response is coming from the machine based on the questions and responses.

Describe game theory. What role does it play in AI?

The scientific and logical field of study known as "game theory" creates a model of potential interactions among two or more reasonable players. Thus, the term "rational" refers to a player's belief that other players share his or her sense of reason and comprehension. In a multi-agent setting, where players must make decisions from a set of possibilities, game theory says that each player's decision has an impact on the other players' decisions.

AI and game theory are closely connected and complementary fields. When several agents are attempting to engage with one another to accomplish a goal in a multi-agent environment, game theory is frequently employed in AI to provide some of the important features needed.

Some well-known games, including Poker and Chess, are logical games with predetermined rules. Algorithms must be developed specifically for these games in order to play them online or digitally on devices like laptops, mobile devices, etc. And to apply these algorithms, artificial intelligence is applied.

An artificially generated neural network is what? Describe a few popular artificial neural networks.

Artificial neural networks, often known as AI neural networks, is statistical constructions that were created by modelling the behaviour of neurons in the human brain. Many AI techniques, including machine learning and deep learning, are used in these neural networks.

There are several layers that make up an artificial neural network, or ANN, including the input nodes, output units, and hidden layers.

What exactly is a chatbot?

A chatbot is an artificially intelligent piece of software or agent that mimics human communication using natural language processing. Via a website, software, or messaging app, you may engage in a chat. These chatbots, often known as digital assistants, can communicate with people verbally or via text.

The majority of organisations utilise AI chatbots, such the Vainubot and HDFC Eva chatbots, to give their clients virtual customer assistance around-the-clock.

How does AI represent knowledge?

The area of artificial intelligence that is concerned with how AI agents think is known as knowledge representation. It is used to provide real-world knowledge to AI agents so they're able to comprehend and make use of it when tackling challenging AI challenges.

Why is a particular programming language not frequently used in AI?

Although it is a programming language, Perl is not a frequently used language for AI.

What does reinforcement learning imply exactly?

Reinforcement learning is a component of machine learning. In this, each agent creates behaviours to interact with its surroundings and learn from feedback. The agent gets feedback in the form of incentives, such as positive incentives for good behaviour and negative incentives for bad behaviour. No color information or supervision are given to the agent. In real life, the agent continuously engages in three activities to study the environment: acting, changing states, and receiving feedback.

The following components form the RL-based system mainly:

Environment: The agent's immediate surroundings are his area of investigation and action.

Agent: An agent is an artificial intelligence (AI) programme with sensors, actuators, and the capacity to perceive its surroundings.

State: The environment responds to the agent with the circumstance.

After completing each action, the agent is rewarded with feedback.

In real life, the agent engages in various actions in order to study the world. Depending on the sort of activity, the agent's status changes or occasionally stays the same, and he receives a reward. Feedback is the prize, and depending on the activity, it may be either favourable or negative.

The agent's objective is to solve the problem and maximise the positive reward.

How is tensor flow employed in artificial intelligence?

The Google Brain team's open-source library framework is called Tensor Flow. It is a math library that is applied in a variety of machine learning applications. Tensor flow makes it simple to deploy and train machine learning algorithm in the cloud.

Which face recognition algorithm does Facebook use? Explain how it operates.

When you submit a photo to Facebook, the DeepFace feature, which employs deep learning techniques for facial identification, offers you options for photo tags. Using neural network models, the deep face recognises faces in digital photos. The following stages describe how DeepFace functions:

The uploaded images are initially scanned. It creates a 3-D representation of the image before rotating it into various positions.

It then begins matching after that. It uses a model of neural networks to find the most striking similarities among various images of a person in order to match that one. It looks for several characteristics like the distance between both the eyes, the curvature of something like the nose, the colour of the eyes, etc.

Following that, it performs the recursive testing for 68 landmarks, since each human face has 68 distinct facial points.

Following mapping, it encodes that image and looks up the person's information.